The NVDIMM/DAX pull for Linux 7.3 contains preparatory work for FAMFS. That is different from merging the filesystem itself. FAMFS has a separate v13 proposal aimed at file-based access to very large shared or disaggregated memory.
Two different pieces of kernel work
The DAX pull request lists fixes to mapping offsets, failure handling and device ownership. Its merge confirmation is dated August 20. It does not add the FAMFS filesystem.
John Groves’s separate v13 series was submitted on August 10. Calling FAMFS “not submitted” is therefore misleading: the code was posted for review, while the prerequisite merge and acceptance of the filesystem are distinct events. Review activity does not guarantee a future acceptance date.
Files name the shared bytes
The proposal exposes memory through files that applications can map directly, without using the page cache for the data path. Multiple nodes can mount it if they have access to the underlying memory. Its stated scope is specialized shared-memory appliances, not a replacement for a normal root filesystem.
For an original example, suppose host A and host B both map a named dataset backed by the same accessible memory region. Mapping the file gives each application a way to address the data. It does not, by itself, establish when a new version is complete or whether both applications may overwrite it at once.
An application still needs a publication and synchronization protocol appropriate to the actual hardware and software. A reader seeing a “ready” flag before the corresponding data is visible would be incorrect even though the file opened successfully. The filesystem name alone does not prove cross-host cache coherence.
Direct access is not a durability promise
Memory mapping, sharing and survival after power loss are separate properties. A pool backed by volatile memory does not become persistent merely because the bytes have filenames. Check the device’s failure and persistence model, along with the application’s recovery procedure.
For an evaluation, keep three questions separate: can both nodes reach the intended region, does the application observe correctly synchronized data, and what survives the failures you care about? Avoid formatting an ordinary disk or migrating a root filesystem on the assumption that FAMFS is a general upgrade. This article reviews the submitted design; it does not report a PeopleAreGeek CXL deployment.
Separate merged DAX prerequisites from submitted FAMFS v13; avoid claiming universal CXL coherence, persistence or guaranteed acceptance.